Fine-grained classification on Aircraft (Clean Accuracy, Adversarial Accuracy)
25Clean AccuracyMTA
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MTABackbone=CLIP-ViT-B/16, Perturbation bound (epsilon)=4.02025.12 | 25 | 4.3 | |
| CLIPBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 24 | 0 | |
| CLIPBackbone=CLIP-ViT-B/16, Perturbation bound (epsilon)=4.02025.12 | 23.9 | 0 | |
| R-TPTBackbone=CLIP-ViT-B/16, Perturbation bound (epsilon)=4.02025.12 | 23.9 | 13.2 | |
| TTPBackbone=CLIP-ViT-B/16, Perturbation bound (epsilon)=4.02025.12 | 23.9 | 14.8 | |
| EnsembleBackbone=CLIP-ViT-B/16, Perturbation bound (epsilon)=4.02025.12 | 23.4 | 8.7 | |
| TTCBackbone=CLIP-ViT-B/16, Perturbation bound (epsilon)=4.02025.12 | 23.3 | 0.5 | |
| EnsembleBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 23.3 | 4.1 | |
| SS-TPTBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 22.5 | 12.8 | |
| R-TPTBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 21.7 | 11.7 | |
| TPTBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 21.6 | 0 | |
| DOCBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 21.2 | 0.2 | |
| TTCBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 20.8 | 0.3 | |
| MTABackbone=CLIP-ViT-B/32, Perturbation Bound (epsilon)=4.02025.12 | 20.2 | 8.2 | |
| TAPTBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 20.1 | 2.9 | |
| R-TPTBackbone=CLIP-ViT-B/32, Perturbation Bound (epsilon)=4.02025.12 | 19.1 | 9.2 | |
| TTCBackbone=CLIP-ViT-B/32, Perturbation Bound (epsilon)=4.02025.12 | 18.2 | 1 | |
| MTABackbone=CLIP-ResNet50, Epsilon=1.02025.04 | 18.1 | 3.7 | |
| CLIPBackbone=CLIP-ViT-B/32, Perturbation Bound (epsilon)=4.02025.12 | 18.1 | 0 | |
| TTPBackbone=CLIP-ViT-B/32, Perturbation Bound (epsilon)=4.02025.12 | 18 | 10.3 | |
| R-TPTBackbone=CLIP-ResNet50, Epsilon=1.02025.04 | 17.5 | 12.6 | |
| TPTBackbone=CLIP-ResNet50, Epsilon=1.02025.04 | 17.3 | 0 | |
| C-TPTBackbone=CLIP-ResNet50, Epsilon=1.02025.04 | 16.7 | 0 | |
| EnsembleBackbone=CLIP-ResNet50, Epsilon=1.02025.04 | 16.4 | 9.8 | |
| EnsembleBackbone=CLIP-ViT-B/32, Perturbation Bound (epsilon)=4.02025.12 | 16.4 | 7.9 | |
| CLIPBackbone=CLIP-ResNet50, Epsilon=1.02025.04 | 15.7 | 0 | |
| CLIPBackbone=CLIP-ViT-B/32, Pre-trained=TeCoA, Augmented views=15, Epsilon=4.02026.06 | 6.6 | 0.7 | |
| TPTBackbone=CLIP-ViT-B/32, Pre-trained=TeCoA, Augmented views=15, Epsilon=4.02026.06 | 6.6 | 1.2 | |
| SS-TPTBackbone=CLIP-ViT-B/32, Pre-trained=TeCoA, Augmented views=15, Epsilon=4.02026.06 | 5.5 | 2.6 | |
| R-TPTBackbone=CLIP-ViT-B/32, Pre-trained=TeCoA, Augmented views=15, Epsilon=4.02026.06 | 5.4 | 2.2 | |
| EnsembleBackbone=CLIP-ViT-B/32, Pre-trained=TeCoA, Augmented views=15, Epsilon=4.02026.06 | 4.4 | 1.9 |